利用無人機多光譜影像推估百慕達草坪施肥管理之反應研究

Research on Unmanned Aircraft Vehicle Multispectral Images to Estimate Fertilization Management in Bermudagrass Turf

張博緯(Chang Po-Wei)

指導教授 ： 謝清祥

國立屏東科技大學/國際學院/熱帶農業暨國際合作系/碩士(2019年)

https://doi.org/10.6346/NPUST201900324

摘要

無人飛行器（UAV）近年來成為最有利於監測作物健康的技術，其技術與發展依循GPS和大數據分析來獲取更多資訊。在台灣，草坪在運動領域的使用正逐漸成長。草坪品質是評估管理操作中草坪性能的表現。它的組成主要是由顏色，密度，葉片質地，生長習性，平滑度和均勻性等六大項品質表現。
百慕達草在台灣被利用在高爾夫球場及運動場較多，其中雜交種Tifway419為最常利用之品種，而Tifdwarf 或 Tifgreen等可以被割刈至超矮高度的雜交種被利用在高爾夫球場的果嶺為多。在庭院或者是公園等常可以發現普通種或台灣馴化種百慕達具有較長的葉片及較寬的寬度，在經過割刈管理後可以形成品質中等的草坪。
從前台灣草坪發展與品質監測系統缺乏，而如今精準農業為農藝學帶來了更多的好處，在無人飛行器（UAV）的遙感探測中使用多光譜反射訊號可偵測植物之生長品質。植物捕獲太陽輻射中之可見光譜區段，因為它對生存至關重要。因此，我們從植物的光譜反射中捕獲訊息。在植物中吸收的大部分能量仍然是低於380nm波長的紫外線及在380nm和780nm之可見光。多光譜相機或相關儀器應用光譜反應之不同植生指數來建立單一數值並評估植物生長條件，環境逆境和葉面積成長等狀況。
本試驗之目的在利用UAV多光譜之各項植生指標對比草坪不同肥料下之生長與品質以測試UAV應用之可行性。
在第一次和第二次肥料試驗中(分別為沒有割草和割草的操作)。主要目的在測試四種不同肥料的持續時間。在試驗I(沒有割草)，顏色指數顯示在第1天施肥後色澤增加，直到第29天。而在第4週，結果四種不同的處理無顯著差異。其中在前三週中M處理(N:P:K ; 21%: 3%: 21%)相比L處理(2.5%: 0.5%: 0.5%)和O處理(1.8%: 0.5%: 0.5%)相比有顯著性差異，但與F處理(1.25%: 0.5%: 0.5%)無差異。而第4週的比較中均顯示四種不同處理方法無生長差異。
在試驗II(加入割草操作)結果顯示: 在第2週之後在視覺評估的顏色、顏色指數、綠覆蓋百分比、均一性還有割刈鮮重上皆有顯著差異；M處理在第二週後肥料釋出，顯示出與其他處理之間不同的效果，尤其是在第4週，處理之間有顯著差異。所有視覺評估項目在M處理上都顯示出最佳結果，除了葉子寬度。而其結論是肥料可以保留到第4週且有著顯著差異。並將這些結果應用於試驗3和4。
在試驗III進行了田間試驗(沒有割草操作)，結果顯示在第3週，L處理具有顯著最高的NIR和Red反射率（38.6％和9.99％）。在視覺評級中，M處理顯示與L和F處理相比具有顯著差異，其具有最高的顏色值（7），均勻性（71.11％）和密度（31.07/ 25cm2）。UAV之植生指數在第2週和第3週較敏感。結果中僅顯示SR（Simple Ratio），GCI（Green Chlorophyll Index），NDVI（Normalize Difference Vegetation Index）和GNDVI（Green Normalize Difference Vegetation Index）在處理之間存在顯著差異。不同植生指數的相關性檢測結果，SR與均勻度（r = 0.98）、顏色（r = 0.89）、密度（r = 0.71）、鉀（r = 0.91）、葉綠素a（r = 0.90）、葉綠素總量（r = 0.84）、類胡蘿蔔素（r = 0.87）、和花青素（r = -0.9）相關性最高。綠色覆蓋與NDVI的相關性最高（r = 0.88）。
在實驗IV(加入割草操作)，結果顯示NIR反射率在第3週時最顯著，其中且M處理顯示與F處理的差異(37.06 v.s.33.83%)。M處理在紅光上具有顯著最低之反射率(8.67%)。第3週NDVI的差異在於M和L處理（均為0.62）對於O處理（0.58），顏色指數M，L處理（6.26, 6.24）對於O處理（5.98）。視覺評估在第1週和第2週各處理間明顯差異。顏色和綠色覆蓋的顯著差異在於M和L處理，顏色（8.56 v.s. 6.89），綠色覆蓋（78.89 v.s.70％），密度（22.6 v.s.20.13 shoots/ 25cm2），鮮重（116.99g v.s.48.01g）。在第2週各元素含量處理間呈顯著差異: M和L處理差異最大，氮（3.46v.s. 3.12％），磷（0.42 v.s. 0.34％），鉀（2.07 v.s. 1.73％）。色素含量中M和F處理間有顯著差異，葉綠素a （0.11 v.s. 0.10µg/ml），總葉綠素（0.13 v.s. 0.12µg/ml），類胡蘿蔔素（10.85 v.s. 10.38µg/ml）和花青素（0.15 v.s. 0.13µg/ml）。
進一步分析UAV多光譜之植生指數與草坪品質之相關性，結果顯示: NDVI與均勻度（r = 0.85）、氮（r = 0.63）、磷（r = 0.90）和葉綠素a（r = -0.52）的相關性最大。GCI與Green Cover（r = 0.8）和Color（r = 0.92）的相關性最高。Red與雜草百分比最相關（r = -0.81）。GRVI的葉綠素b最多（r = -0.51），花青素（r = -0.55）。MSR對鮮重的影響最大（r = 0.88）。
總結，利用UAV多光譜之各植生指數對比實際草坪生長品質調查，NDVI和SR或MSR對於視覺評估和元素，色素含量具有最大的相關性和敏感性。NDVI具有最大的檢測不同條件的能力。但SR和MSR可以與NDVI共同使用來增加應用上的可靠度。

關鍵字

植生指數 ； 肥料 ； 無人機 ； 多光譜 ； 百慕達草坪

並列摘要

Unmanned aerial vehicles (UAVs) is the most advantage technology that provides benefits to monitor crops health. The Drone (UAV) development was following by GPS and Big Data analysis. In Taiwan, use of turfgrass is rising in sport filed. Turf quality is an evaluation to determine the turfgrass performance under culture management practice. It is performance of color, density, texture, growth habit, smoothness and uniformity in the turf.
Bermudagrass is commonly used in golf courses and sport fields in Taiwan. The hybrid species Tifway419 is the most commonly used variety, while the Tifdwarf or Tifgreen and ultra-short height hybrids are mostly on the golf green. In the courtyard or in the park, it is common to find common bermudagrass or Taiwanese domesticated species. Common bermudagrass has longer and wider blades, and can form a medium-quality lawn after cutting and management.
Monitoring system on Turfgrass in Taiwan is rarely conducted. Nowadays, precision agriculture brings more benefit to the agronomy studies, especially in the remote sensing UAV. The plant captures solar radiation on visible light region which is essential to plant survival. And, UAV uses the reflectance on plants to capture the information and analyze its conditions. Most light energy absorbed in plant in ultraviolet region that is below 380 nm wavelength, and 380 nm to 780 nm is known as visible light. The application of vegetation indices are through the use of multispectral camera or instrument which creates a single light reflecting value to evaluate plant growth condition, environmental stress, and leaf area. The purpose of this experiment was to test the feasibility of UAV application by using the UAV multi-spectral plant indices to compare the growth and quality of turf under different fertilizers.
In the first and second fertilizer experiment the main purpose was to test fertilization duration while (turf under no mowing and mowing practice) under four different fertilizers. In the first study (no mowing practice), the color index showed increased after Day 1 fertilizer application and last until the Day 29. At week 4, the results didn’t show significant differences among four different treatments. Before week 4, M treatment performed significant difference as compared to L and O treatment, but no difference to F treatment. The comparisons on week 4 all growth characters showed no difference in four different treatments.
In the second study (mowing practice) the result showed significant difference on Week 2 and weeks after. M treatment showed the huge effect compared to other treatments, especially in week 4. All visual rating characters showed the best results on M treatment, except leaf texture, which also concluded the fertilizer can retained to week 4. And, this information was combined and used on experiment III and IV.
In experiment III(no mowing practice) on field test, the results showed significant difference on week 3. L treatment had highest NIR and Red reflectance (38.6% and 9.99%)；In the visual rating, M treatment showed the significant difference compared to L and F treatment, which had the highest value on color (7), uniformity (71.11%), and density (31.07 /25cm2). Vegetation indices produced by UAV was sensitive on week 2 and week 3. The results showed that SR (Simple Ratio), GCI (Green Chlorophyll Index), NDVI (Normalized difference vegetation index), and GNDVI (Green Normalized difference vegetation index) had significant difference among treatments. The correlation for different vegetation indices to grow characters showed SR were highly related to uniformity (r=0.98), color (r=0.89), density (r=0.71), potassium (r=0.91), chlorophyll a (r=0.9), total chlorophyll (r=0.84), carotenoid (r=0.87), ad anthocyanin (r=-0.9). And the green cover showed high correlation on NDVI (r=0.88).
In the experiment IV field test (mowing), NIR reflectance had most significant on week 3, M treatment showed the difference compared to F treatment (37.06 vs. 33.83%). M treatment had significant lowest reflectance on Red (8.67%). At Week3 NDVI showed significant differences between M and L (both 0.62) with O (0.58), and color index M, L (6.26, 6.24) with O treatment (5.98). The visual rating had significant difference among treatments at week 1 and week 2. The significant difference on color and green cover was between M and L treatment, color (8.56 vs. 6.89), green cover percentage (78.89% vs. 70%), density (22.6 vs. 20.13 shoots/25cm2), and mowing fresh weight (116.99g vs. 48.01g). At week 2, element contents showed significant difference among treatments；M and L showed most significant difference on nitrogen (3.46% vs. 3.12%), phosphorus (0.42% vs. 0.34%), and potassium (2.07% vs. 1.73%). On pigment contents, M and F had significant difference on chlorophyll a (0.11 vs. 0.10µg/ml), total chlorophyll (0.13 vs. 0.12µg/ml), carotenoid (10.85 vs. 10.38µg/ml), and anthocyanin (0.15 vs. 0.13µg/ml).
Correlation coefficient on UAV vegetation Index combined turfgrass visual quality rating showed that NDVI had significant correlation to uniformity (r=0.85), nitrogen (r=0.63), phosphors (r=0.90), and chlorophyll a (r=-0.52). GCI showed highest correlation to green cover (r=0.8), and color (r=0.92). Red was most correlated on weed percentage (r=-0.81). GRVI was most correlated to chlorophyll b (r=-0.51), anthocyanin (r=-0.55) ；MSR had highest correlation to fresh weight (r=0.88).
In conclusion, the results on comparison of UAV vegetation indices with visual turfgrass quality evaluation showed that NDVI and SR or MSR had most correlation and sensitivity to the visual quality, element, and pigment contents. The NDVI had the most ability to detect different plant conditions. However, SR and MSR can be used together with NDVI to increase application reliability.

並列關鍵字

Vegetation Index ； fertilizer ； unmanned aerial vehicle ； multispectral ； Bermudagrass turf

參考文獻

Arai, K., K. Gondoh, O. Shigetomi, and M.Yuko. 2016. Method for NIR Reflectance Estimation with Visible Camera Data Based on Regression for NDVI Estimation and Its Application for Insect Damage Detection of Rice Paddy Fields. International Journal of Advanced Research in Artificial Intelligence 5(11):17–22.

Arnon, D. I. and F. R. Whatley. 1954. Metabolism of Isolated Cellular Particles from Photo- Synthetic Tissues I. Oxygen Uptake and Carbon Dioxide Evolution in the Dark. Physiologia Plantarum 7:602–13.

Baghzouz, M., D. A. Devitt, and R. Morris. 2007. Assessing Canopy Spectral Reflectrance of Hybrid Bermudagrass under Various Combinations of Nitrogen and Water Treatments. Applied Engineering in Agriculture 23:763–74.

Bannari, A., H. Asalhi, and P. Teillet. 2002. Transformed Difference Vegetation Index (TDVI) for Vegetation Cover Mapping. Canada Centre for Remote Sensing 3053–55.

Barati, S., B. Rayegani, M. Saati, A. Sharifi, and M. Nasri. 2011. Comparison the Accuracies of Different Spectral Indices for Estimation of Vegetation Cover Fraction in Sparse Vegetated Areas. Egyptian Journal of Remote Sensing and Space Science 14(1):49–56.